That part is kind of weird to me. I understand you want someone to be "responsible", but what do you expect the response to be? This doctor failed to deliver, let's fire himher. What follows? You get a better doctor, hopefully. Wouldn't it be better to have an algorithm that can be improved when it fails, so you don't have to train a whole new doctor to replace the bad one? Plus, you can have every single robot doctor be exactly as good at everything. No more "you need to find the best surgeon for X". You can have as many best surgeons as you can economically build.
The responsibility would fall on the robot's designers to improve their software, so it doesn't make the mistake in question. And you will be able to assure people that no doctor will ever make that mistake ever again. Imagining that all these robots are linked, or at least synchronise every couple of days, they would learn a lot faster than a human.